Nathan Hartman wrote on Sun, Nov 17, 2019 at 00:16:56 -0500:
> (but all the recent failures are for the same reason):
> > Updating '.':
> > svn: E170013: Unable to connect to a repository at URL '
> > svn: E120108: Error running context: The server unexpectedly closed the
> What's with all these failures lately?
Ask Infra to check the server's error log? They'll want to know the timestamp,
the IP of the host the build ran on (if we have it; otherwise, the buildbot
builder's name, whence they can presumably work out the IP via the buildmaster),
and that the error was from svn-master.a.o (as opposed to a mirror).
> Is there a reason the buildbot needs to use
> as opposed to
Yes. "svn.apache.org" has several mirrors, using the DAV proxy feature. The
scenario behind this is:
- Some other Apache project commits a 3000-file javadoc update in rN
- We commit something in rN+1
- The buildbot slave updates to rN+1 before rN finishes svnsync'ing over to the
svn.a.o mirror used by that buildbot slave
If the buildbot slaves used https://svn.a.o, they'd get the "Revision does not
exist" error when they tried to update/checkout their working copies. Therefore,
they get their working copies from the master (behind the DAV proxy) directly.
This is ultimately because /repos/asf is a single repository, svnsync is sequential,
and the DAV proxy abstraction gives "Revision does not exist" for revisions that haven't
been svnsynced from the master.
> Or should it be svn-master but with https?
That can't hurt. You can change this yourself by editing
(It'll get auto-deployed within five minutes of the commit. You can use
źpython -mpy_compile╗ to sanity check the diff before you commit it.)
Received on 2019-11-17 14:39:11 CET